FAIR: Frequency-aware Image Restoration for Industrial Visual Anomaly Detection

Abstract

Image reconstruction-based anomaly detection models are widely explored in industrial visual inspection. However, existing models usually suffer from the trade-off between normal reconstruction fidelity and abnormal reconstruction distinguishability, which damages the performance. In this paper, we find that the above trade-off can be better mitigated by leveraging the distinct frequency biases between normal and abnormal reconstruction errors. To this end, we propose Frequency-aware Image Restoration (FAIR), a novel self-supervised image restoration task that restores images from their high-frequency components. It enables precise reconstruction of normal patterns while mitigating unfavorable generalization to anomalies. Using only a simple vanilla UNet, FAIR achieves state-of-the-art performance with higher efficiency on various defect detection datasets. Code: https://github.com/liutongkun/FAIR.

arXiv comment12 pages, 10 figures

References in corpus (15)

• Denoising Diffusion Probabilistic Models
• Diffusion Models in Vision: A Survey
• Gradient Magnitude Similarity Deviation: A Highly Efficient Perceptual Image Quality Index
• Uninformed Students: Student-Teacher Anomaly Detection with Discriminative Latent Embeddings
• Deep Industrial Image Anomaly Detection: A Survey
• Deep Anomaly Detection Using Geometric Transformations
• Using Self-Supervised Learning Can Improve Model Robustness and Uncertainty
• Omni-frequency Channel-selection Representations for Unsupervised Anomaly Detection
• FastFlow: Unsupervised Anomaly Detection and Localization via 2D Normalizing Flows
• Deep Learning for Unsupervised Anomaly Localization in Industrial Images: A Survey
• DFR: Deep Feature Reconstruction for Unsupervised Anomaly Segmentation
• Learning and Evaluating Representations for Deep One-class Classification
• A Unified Model for Multi-class Anomaly Detection
• Natural Synthetic Anomalies for Self-Supervised Anomaly Detection and Localization
• DiffusionAD: Norm-guided One-step Denoising Diffusion for Anomaly Detection